What architecture layers does AdiOS have?

AdiOS has four architecture layers plus apps. L2 (substrate) — adios-ignition, adios-kernel, adios-flintcore, adios-siliconbridge, adios-terra, adios-formfactor, adios-node. L3 (plugins) — 32 components covering AI inference, analysis, domain twins, integration, marketplace, and tools. L4 (intelligence) — 34 core components including data/knowledge, cognition, governance, orchestration, and compute. L5 (gateway + apps) — adios-gateway, adios-portal, adios-forge, adios-one-ui (React+Tauri), adios-one (Leptos WASM).

What languages and runtimes does AdiOS use?

AdiOS is written 100% in Rust, end-to-end. Kernel primitives, runtime, knowledge graph, governance, inference gateway, and even the frontend (Leptos WASM reference + React+Tauri primary) are all Rust. No Python, no Node runtime in production. The Rust monoculture gives type-checked compliance, no interpreter surprises, and source-auditability for sovereign deployments.

What hardware does AdiOS support?

AdiOS runs on x86_64, ARM64 (aarch64), RISC-V, NVIDIA (CUDA), and Apple Silicon (Metal). Hardware abstraction lives in adios-siliconbridge. The platform ships 10 form-factor manifests via adios-formfactor: HPC, server, edge, FTD, citizen, SME, gateway, kiosk, mobile, and IoT — one codebase across all of them.

How many components and tests does AdiOS have?

As of 2026-05-09: 90 Rust components (53 stable, 32 beta, 5 alpha), 7,400+ passing tests across 734 source files, 125 Architecture Decision Records (latest ADR-127), 96 spec directories, 318,000+ lines of code. Continuously verified by `cargo xtask validate-catalog` against the canonical manifest at `.adios-platform/config/service_catalog.yaml`. Notable test counts: adios-meridian 592 (CRDT), adios-mirror 193 (twin engine), adios-gateway 123 (with catalog loader), adios-regionkit 106 (13 regulators), adios-seedvault 113, adios-fhir 58 (FHIR R4 + ABDM + EHDS), adios-pharma 43 (cGMP + 21 CFR Part 11).

What ports do AdiOS HTTP services use?

AdiOS uses ports 7878 (forge) and 8440-8471 for 16 HTTP services: 8440 seedvault, 8441 identity, 8442 graphcore, 8443 gateway, 8444 cortex, 8445 neuralmesh, 8446 governance, 8447 fabric, 8448 sentinel, 8449 exchange, 8450 bfsi-server, 8452 scribe, 8460 planner, 8470 mirror, 8471 inference. Port allocation is documented at https://www.adiosplat.io/service-catalog.html .

Is AdiOS open source?

AdiOS is sovereign-first commercial software with select components available open source. Open-source components include adios-seedvault and adios-meridian. Enterprise deployments are licensed per-division with on-premises, private-cloud, and managed options. Patent claims (9 total, Chennai Patent Office priority March 2026) protect the Sovereign Knowledge Graph, Confidence-Tiered Promotion, CRDT Offline Sync, Intent-Driven Provisioning, Universal Form Factor, Sovereign Lakehouse Zero-Copy IPC, Policy-Ring Governance, Circular Knowledge Compounding Loop, and Cognitive Architecture.

Cloud AI extracts.
AdiOS compounds.

AdiOS is India's sovereign AI Operating System — the building. COS is the architectural blueprint inside it. This page is about the building.

90 Rust components across 4 architecture layers + apps. 9 patent claims filed March 2026. DPIIT-recognised Deep Tech Startup.

86
Components

9
Patents

318K
LOC Rust

7,400+
Tests

13
Regulators

81
ADRs

Explore the OS Book a Call

adios boot | sovereign init

$ adios init --sovereign --mode=enterprise
> mounting brain.kernel v1.4.2 ............ ✓
> loading lakehouse.engine ................ ✓
> activating sentinel.enforcement ......... ✓
> 5,000+ tests passing .................... ✓
> confidence.baseline: 0.70 → 0.95+ ....... ✓
> query.latency: 127ms avg ................ ✓
> data.egress: ZERO ...................... ✓

SOVEREIGN MODE ACTIVE. Boot complete._

Act I

The Operating System

Every generation of computing has been defined by its operating system. DOS gave us the command line. Unix gave us multi-user computing. Windows gave us the GUI. Linux gave us the server. Android gave us the pocket. Each era's OS was not just software. It was the substrate of possibility.

We are at the threshold of a new era. Intelligence is moving from the cloud to the kernel. From external service to foundational layer. The question is not whether AI will be embedded into enterprise infrastructure. It is whether that intelligence will be sovereign or rented.

"What if intelligence was as fundamental as the file system?"

adios.system_tray LIVE

brain.status ● active | tier 3 enterprise

lakehouse.status ● connected | 847 entities indexed

sentinel.status ● enforcing | 12 active policies

query.latency ◆ 127ms avg | confidence: 0.94

sovereignty ● local | zero cloud egress

	Traditional AI	With AdiOS
Data Sovereignty	Data leaves India	Data never moves
Provenance	0% traceability	100% traced
Confidence Scoring	None \| black box	0.70 → 0.85 → 0.95+
Compliance	Manual \| 3 weeks	Automated \| inline
Query Speed	7 months manual	127 milliseconds
Breach Surface	500+ TB exposed	0 TB \| zero movement

Act II

The Three-Layer Brain

Intelligence without trust is noise. The AdiOS Brain structures knowledge across three tiers. Each assertion earns its confidence score through review, validation, and provenance.

Tier 1

0.70

Personal

Individual assertions and context. Starting point for all knowledge.

Peer review gate

Tier 2

0.85

Domain

Department-scoped knowledge. Reviewed by domain experts.

Expert validation gate

Tier 3

0.95+

Enterprise

Validated, cross-domain, high-confidence truth.

ENTERPRISE TRUTH

"Trust is earned, not declared."

Act III

Three Pillars

Every intelligent query touches three foundational systems simultaneously. Not sequential. Concurrent. This is what makes AdiOS architecturally distinct.

WHO WHAT WHY

Brain

The knowledge graph and trust layer. Three-tier hierarchy from personal assertion to enterprise truth. Every fact carries confidence and provenance.

NeuralMesh | Oxigraph + RocksDB | SPARQL

WHEN WHERE HOW MUCH

Lakehouse

Unified data fabric. Time-series, geospatial, and quantitative context. All data remains local. Zero cloud egress, full provenance chain.

Strata | DuckDB + Apache Iceberg | SQL

WHO MAY SEE WHAT

Governance

Sentinel enforcement layer. Policy-driven access control with audit trails. Sovereign from day one. No data leaves without explicit, logged permission.

Progressive Identity: Standard (JWT/OAuth) by default. DID-Enhanced and Full SSI available as opt-in layers for cross-enterprise trust.

Sentinel + RegionKit | SHACL + OPA | W3C DID/VC

127 milliseconds. Confidence-scored. Provenance-attached.

Act IV

Architecture

Five divisions. Forty components. One coherent system. Designed from first principles, not assembled from parts.

15 Core

the heart — graph, brain, governance, data

GraphCore DeepRoot Cortex Sentinel Conductor Catalog Lineage Assay SeedVault NeuralMesh Prism Meridian RegionKit Strata DataFS

5 Substrate

where it runs — OS, kernel, hardware

Ignition FlintCore Terra Kernel SiliconBridge

7 Tools

how users interact — explorer, studio, portal

Forge Lens Grid Query Blueprint Kanban Portal

6 Integrations

how it connects — connectors, APIs, ingestion

DataFlow Jupyter MCP Plugins SAP Gateway

7 Ecosystem

how it grows — AI, analysis, marketplace

Inference Domain AI Scout Radar Compass Mirror Exchange

0 tests passing

Act V

Three capabilities most platforms can't claim.

Beyond the headline OS layer, AdiOS ships shipped-and-tested coverage of three areas that rarely live in the same codebase: HPC, Compliance, and Developer Builder OS.

→ Want to see how it actually works? Walk the interactive architecture tour — 5 layers, 6 invariants, 90 components, and the circular compounding loop (Patent S8) traced live. Two minutes. Zero LLM tokens. Horizontal sector switcher + vertical stack visualisation.

SOVEREIGN HPC

Above PARAM. Above NSM. Above AIRAWAT.

India runs world-class supercomputing hardware on imported software. adios-mpi + adios-jobs + adios-siliconbridge close the gap. Not a competitor — the missing OS layer.

Slurm + PBS submit, sentinel-gated
NVIDIA + Apple Metal + CPU silicon abstraction
Air-gap friendly via Meridian CRDT
NSM Phase 3 / AIRAWAT positioning (ADR-078)

HPC sector on Radar →

COMPLIANCE OS

13 regulators. India + EU. One codebase.

421 automated controls live in adios-regionkit, enforced at every cross-component call by adios-sentinel. Customers in Mumbai and Munich on the same platform.

India: DPDP, RBI Free-AI, SEBI, IRDAI, ABDM, CDSCO, Schedule M
EU: GDPR Art 9, EHDS, MDR, IVDR, EU GMP Annex 11
FHIR R4 + EHDS profiles in adios-fhir (58 tests)
21 CFR §11 + cGMP in adios-pharma (43 tests)

Pharma sector on Radar →

BUILDER OS

IDE, designer, query builder, marketplace.

Every component a developer needs to ship sovereign workloads — bundled. Code, design, query, and publish from one platform.

adios-forge — IDE service (port 7878)
adios-canvas — visual COS designer
adios-query — SPARQL builder
adios-exchange — sovereign marketplace (port 8449)
adios-scribe — content engine (port 8452)

Full service catalog →

AdiOS vs Alternatives

How AdiOS compares to existing approaches across the dimensions that matter for sovereign enterprise AI.

Capability	AdiOS	Cloud AI	Local LLMs	Data Catalogs	Compliance Tools
Data Sovereignty	✓ On-prem	✗ Cloud-only	✓ Local	✗ Cloud metadata	✗ Cloud
Semantic Knowledge Graph	✓ Full SPARQL	✗ No	~ Embeddings	~ Partial	✗ No
Confidence Scoring	✓ 9 patents	✗ No	✗ No	✗ No	✗ No
Machine-Executable Compliance	✓ Architecture-native	✗ Manual	✗ Manual	✗ Manual	~ Config-based
Offline-First	✓ Full offline	✗ Cloud-dependent	✓ Local	✗ Cloud-dependent	✗ Cloud-dependent
Unified Query	✓ SPARQL + SQL	~ SQL only	~ LLM only	~ Metadata	✗ No
OS-Level Integration	✓ Kernel level	✗ App layer	✗ App layer	✗ App layer	✗ App layer

AdiOS vs named competitors

The 2026 competitive map across three layers: global "AI OS" / control plane, AI-native data substrate, and Indian AI-OS plays. We don't compete with everyone — many are partners or integrators. The table below is the candid view.

Layer	Competitor	Their strength	AdiOS posture
L1 GLOBAL	Palantir AIP	Gold standard for US enterprise / DoD AI control plane. Mature ontology + workflow.	Don't fight on US turf. Win on India sovereignty + 1/10th cost + DPDP/RBI native.
L1 GLOBAL	NVIDIA (NIM, Dynamo, DGX)	GPU + CUDA + AI runtime ecosystem. Heavy India investment.	Partner. AdiOS is the OS above NVIDIA hardware. Joint sovereign-stack pitch.
L1 GLOBAL	Hyperscalers (AWS Bedrock, Azure OpenAI, GCP Vertex)	Massive reach, $20B+ Indian DC investment. Easy if you're already cloud-native.	CLOUD Act exposure remains. Win regulated workloads (BFSI, healthcare, govt) where on-prem is the procurement gate.
L1 GLOBAL	Databricks + Snowflake	"AI Data Cloud" climbing up-stack. Strong analytics + ML maturity.	Live at L4/L5 of our stack. Ingest sources via `adios-stream`. Not a head-on enemy.
L2 SUBSTRATE	VAST Data	AI-native disaggregated storage paired with GPU clusters.	Partner. AdiOS is the brain on top. Joint pitch: VAST plumbing + AdiOS sovereignty.
L2 SUBSTRATE	Dell PowerStore · HPE Alletra · Nutanix · Ceph · Scality	Enterprise + open-source storage substrates.	Partners, all of them. AdiOS layers above any of these.
L3 INDIA	Superhealth — SuperOS (healthcare)	Hospital end-to-end AI OS, deployed in own flagship.	Direct competitor in healthcare. We win on regulator-mapping (FHIR + ABDM + EHDS); they win on clinical-workflow depth. Different buyer entry.
L3 INDIA	MahaCrimeOS (Microsoft + CyberEye)	AI OS for police investigations across 1,100 stations on Azure + Copilot.	Pattern to watch. Microsoft will replicate this CityOS/GovOS pattern across Indian states. Sovereign-by-architecture is our counter-pitch.
L3 INDIA	UnifyApps	"AI OS architecture" for SaaS workflows + governance + no-code AI app builder.	Adjacent. UnifyApps = iPaaS+++ ; AdiOS = OS-level (kernel + Sentinel ring + KG). Win when the buyer needs on-prem + sovereign.
L3 INDIA	Sarvam AI	Sovereign Indic LLMs (translation, foundation models). IndiaAI Mission partner.	Partner / model vendor. AdiOS runs Sarvam models. Joint sovereign-stack story.
L3 INDIA	Krutrim (Ola)	Full-stack: own LLMs + DC infra + cloud-style services + NVIDIA partnership.	Partial competitor + partial partner. Krutrim wants the cloud; AdiOS is the OS above. Founder-level dialogue valuable.
L3 INDIA	Uniphore · Yellow.ai · CoRover (BharatGPT)	Conversational + agent platforms. Strong CX / contact-centre traction.	Adjacent / ecosystem. They sit on top of AdiOS. Marketplace partnership candidates.
L3 INDIA	Fractal · Arya.ai · Locus	Decision intelligence + vertical AI services in BFSI / logistics.	Partial overlap. Mostly services-heavy; AdiOS is product-heavy. Different buyer journey.

Internal mantra: Palantir, NVIDIA, and the hyperscalers are the global AI super-highways. AdiOS is the local rail network — an AI Operating System built for India's data, regulations, and workflows, that sits on top of whichever infra you choose.

Cloud AI extracts.AdiOS compounds.